JavaScript, WebAssembly, and module workers are required. Each browser tab runs an independent simulation.
Python · Pyodide · WebAssembly · Hosted on Vercel
Restarts fleet, street and arm. Clears findings, faults and taught virtual poses in this tab.
THE MISSION CONTINUES / SHELLHACKS 2026
Two scouts. One smarter workflow.
Explore, inspect, and hand off evidence. Spend intelligence where it matters.
A software proving ground. Movement, sensors, findings, and model decisions are simulated. This page makes no paid API calls. Arm handoffs are previews.
LOCAL METRIC ENVIRONMENT
Inspection floor
STREET INSPECTION REPLAY
Loading the crawler's street view…
The browser bundle is loading an independent inspection replay with simulated evidence.
ECONOMY BY DESIGN
Every model call should earn its place.
Local sensor checks filter routine readings. New evidence triggers a simulated analysis; repeated evidence stays local.
Costs use illustrative rates. Travel and battery are simulated. No energy, carbon, or environmental savings have been measured.
View cost and token assumptions
Each modeled request uses 1,800 input tokens and 220 output tokens. Illustrative rates are $0.30 per million input tokens and $2.50 per million output tokens. These are scenario assumptions, not provider pricing or measured token usage.
The baseline analyzes every observation, including periodic readings and arrival evidence.
| Workflow accounting | Every observation | Evidence gated |
|---|---|---|
| Model requests | 0 | 0 |
| Tokens | 0 | 0 |
| Illustrative cost | $0.0000 | $0.0000 |
Start a mission to compare the policies against its sensor evidence.
MISSION EVENT TRAIL
The reasoning stays visible.
- Start a mission to follow navigation, sensing, decisions, and recovery.
FROM INSPECTION TO INTERVENTION
Arm handoff preview.
Inspection evidence can become a reviewed arm task. This page runs the virtual arm; physical playback is not verified.
VIRTUAL ARM / SIX-JOINT REHEARSAL
Rehearse the intervention.
Pick and place a virtual marker, teach a routine, and test recovery. This rehearsal runs independently of the scouts.
Marker state: awaiting simulator
PRESET MARKER TRANSFER
A complete routine, ready to play.
Approach the marker, close the gripper, lift, transfer, release, and return home. Grasp and contact are scripted.
Restores the home pose; clears this arm's events and all taught poses.
Connecting…
Test an arm failure
Clearing a fault restores the virtual checkpoint, including a dropped marker, and leaves the arm paused. No physical recovery is performed.
Teach a virtual routine
Set a pose while the arm is idle, record it, and repeat. Replay follows only the poses you taught in this simulator.
- No poses recorded yet.
Arm event trail
BUILT TO REHEARSE, MEASURE, AND EXPLAIN
Technologies in this demo
- Python
- Deterministic fleet missions, obstacle-aware routes, and virtual arm kinematics.
- JavaScript · Canvas 2D
- Interactive floor and arm views, telemetry, and playback controls.
- React · TypeScript · SVG
- The teammate's street replay, animated crawler, and fixture evidence cards.
Recorded local OpenJev inference used typed pollard-jev decisions and synthetic observations. Pollard recorded and allocated model work.
Gemini roles in those trials used fixtures. This simulator runs no model inference; its AI request and cost comparisons are modeled.